The 10% a keyword search of the tariff schedule can never find -- Aervik Labs, measured 2026-09-19
=================================================================================================

WHAT THIS IS. 140 binding classification rulings published by U.S. Customs and Border Protection, in TWO
DISJOINT SAMPLES (A n=100, B n=40) drawn by the same method. For each one we took only CBP's description of the
merchandise and asked where the heading CBP assigned appears in a keyword ranking of every 4-digit heading in
the tariff schedule. No model is involved anywhere in this measurement.

THE RESULT. The correct heading is somewhere in the ranking 96% (A) and 90% (B) of the time.
For 4 rulings in A and 4 in B it scores ZERO -- no depth of list reaches it.

WHAT THAT DOES AND DOES NOT MEAN. It is a measurement of THESE SAMPLES, which are binding rulings: importers
request one when the classification is not obvious, so they are hard cases by construction. It is NOT a claim
that ten percent of all tariff classifications everywhere are impossible with keyword search.

results.csv carries every row: the ruling, CBP's heading, and the rank at which a keyword search finds it.
The schedule is the USITC Harmonized Tariff Schedule snapshot of 2026-09-14.

SOURCE. CBP rulings are works of the U.S. Government and are in the public domain. CBP is cited as the source
and does not endorse this analysis.